Surrogate-Assisted Hybrid Meta-Heuristic Algorithm with an Add-Point Strategy for a Wireless Sensor Network.

Area of Science:

• Computational Intelligence
• Optimization Algorithms
• Machine Learning

Background:

• Meta-heuristic algorithms excel at complex optimization but face challenges with time-intensive fitness function evaluations.
• Surrogate-assisted meta-heuristic algorithms offer a solution for problems with high computational cost per function evaluation.

Purpose of the Study:

• To propose an efficient surrogate-assisted hybrid meta-heuristic algorithm (SAGD) for computationally expensive optimization problems.
• To introduce a novel add-point strategy leveraging historical surrogate model data for improved candidate selection.
• To integrate local Radial Basis Function (RBF) surrogates and a generation-based optimal restart strategy for enhanced performance.

Main Methods:

• Combines the Gannet Optimization Algorithm (GOA) with Differential Evolution (DE) within a surrogate-assisted framework.
• Employs a new add-point strategy using historical surrogate model information to guide true fitness evaluations.
• Utilizes local RBF surrogates for objective function landscape modeling and a control strategy for sample prediction and updates.
• Incorporates a generation-based optimal restart strategy to dynamically select samples for algorithm restarts.

Main Results:

• The proposed SAGD algorithm demonstrated strong performance on seven benchmark functions.
• SAGD proved effective in solving the wireless sensor network (WSN) coverage problem, a representative expensive optimization task.
• The results indicate significant improvements in handling optimization problems with high fitness evaluation costs.

Conclusions:

• SAGD offers an efficient and effective approach for solving computationally expensive optimization problems.
• The novel add-point strategy and integrated components contribute to the algorithm's success in managing high-cost fitness evaluations.
• The study validates the utility of surrogate-assisted meta-heuristics for complex real-world applications like WSN optimization.